A system or method may be used to perform image processing. An example method may include extracting a plurality of sub-images corresponding to a plurality of regions of an image. A rating classification may be generated, using a machine learning trained model, for each of the plurality of sub-images, for example including a no-rating classification for at least one sub-image of the plurality of sub-images. The method may include creating, using the rating classification for each of the plurality of sub-images, a color map overlay of a grayscale image corresponding to the image. The color map overlay may include colors corresponding to respective rating classifications for the plurality of regions or a region without a color corresponding to the at least one sub-image having the no-rating classification. The method may include outputting the color map overlaid on the grayscale image.
Legal claims defining the scope of protection, as filed with the USPTO.
2. The method of claim 1, wherein the rating classifications for each of the plurality of sub-images include a plurality of single value sharpness metrics, and further comprising outputting a matrix including respective absolute values of the plurality of single value sharpness metrics aligned according to an alignment of the plurality of regions of the image.
3. The method of claim 1, wherein the color map overlay corresponds to sharpness within the image.
4. The method of claim 3, wherein the colors of the color map overlay represent absolute values of sharpness.
6. The method of claim 1, further comprising outputting a motion blur rating for the image based on the color map overlay.
7. The method of claim 1, wherein the trained model is trained using a custom training dataset that is generated using limited seed images.
8. The method of claim 1, wherein the trained model includes a convolutional neural network.
9. The method of claim 1, wherein a total number of sub-images used in the plurality of sub-images is user selected.
10. The method of claim 1, further comprising, before generating the rating classification using the machine learning trained model, normalizing and scaling the plurality of sub-images based on training data from training the trained model.
11. The method of claim 1, wherein the color map overlay includes interpolated colors representing rating classification boundaries.
13. The at least non-transitory one machine-readable medium of claim 12, wherein the rating classifications for each of the plurality of sub-images include a plurality of single value sharpness metrics, and wherein the instructions further cause the processor to output a matrix including respective absolute values of the plurality of single value sharpness metrics aligned according to an alignment of the plurality of regions of the image.
14. The at least one non-transitory machine-readable medium of claim 12, wherein the color map overlay corresponds to sharpness within the image.
15. The at least one non-transitory machine-readable medium of claim 14, wherein the colors of the color map overlay represent absolute values of sharpness.
17. The at least one non-transitory machine-readable medium of claim 12, wherein the instructions further cause the processor to output a motion blur rating for the image based on the color map overlay.
18. The at least one non-transitory machine-readable medium of claim 12, wherein the trained model is trained using a custom training dataset that is generated using limited seed images.
19. The at least one non-transitory machine-readable medium of claim 12, wherein the trained model includes a convolutional neural network.
20. The at least one non-transitory machine-readable medium of claim 12, wherein a total number of sub-images used in the plurality of sub-images is user selected.
21. The at least one non-transitory machine-readable medium of claim 12, wherein the instructions further cause the processor to, before generating the rating classification using the machine learning trained model, normalize and scaling the plurality of sub-images based on training data from training the trained model.
22. The at least one non-transitory machine-readable medium of claim 12, wherein the color map overlay includes interpolated colors representing rating classification boundaries.
24. The system of claim 23, wherein the machine learning trained model is trained based on a selection of in-focus real-world source images used to generate sharpness base images that are convolved with a point spread function at a specific single value sharpness metric (SVSM) value and downsampled.
25. The system of claim 24, wherein multiple sub-images are extracted from each convolved downsampled source image and infused with random noise to create a set of augmented training images.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
November 30, 2020
August 16, 2022
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.